Publication number | US7949621 B2 |

Publication type | Grant |

Application number | US 11/871,899 |

Publication date | May 24, 2011 |

Priority date | Oct 12, 2007 |

Fee status | Paid |

Also published as | US20090099990 |

Publication number | 11871899, 871899, US 7949621 B2, US 7949621B2, US-B2-7949621, US7949621 B2, US7949621B2 |

Inventors | Rong Xiao, Xiaoou Tang |

Original Assignee | Microsoft Corporation |

Export Citation | BiBTeX, EndNote, RefMan |

Patent Citations (15), Non-Patent Citations (39), Referenced by (2), Classifications (21), Legal Events (3) | |

External Links: USPTO, USPTO Assignment, Espacenet | |

US 7949621 B2

Abstract

An efficient, effective and at times superior object detection and/or recognition (ODR) function may be built from a set of Bayesian stumps. Bayesian stumps may be constructed for each feature and object class, and the ODR function may be constructed from the subset of Bayesian stumps that minimize Bayesian error for a particular object class. That is, Bayesian error may be utilized as a feature selection measure for the ODR function. Furthermore, Bayesian stumps may be efficiently implemented as lookup tables with entries corresponding to unequal intervals of feature histograms. Interval widths and entry values may be determined so as to minimize Bayesian error, yielding Bayesian stumps that are optimal in this respect.

Claims(20)

1. A method for object detection, the method comprising:

in a processor, implementing the acts of:

instantiating a lookup table for a weak classifier;

setting an entry of the lookup table based on at least a Bayesian error associated with the entry, the entries of the lookup table corresponding to histogram intervals of a feature histogram;

setting a width of the histogram intervals of the feature histogram and a value of a corresponding entry in the lookup table to minimize the associated Bayesian error; and

modifying a computer-readable medium based on at least the weak classifier.

2. The method of claim 1 , wherein the entry of the lookup table is set automatically minimize the associated Bayesian error.

3. The method of claim 1 , wherein:

the entry of the lookup table is associated with at least two histograms; and

the Bayesian error is determined based on at least values in corresponding intervals of the at least two feature histograms.

4. The method of claim 1 , wherein:

the lookup table comprises a plurality of entries; and

the plurality of entries correspond to the histogram intervals of the feature histogram having unequal width.

5. The method of claim 4 , wherein the width of at least one of the histogram intervals is based on at least a distance between the feature histograms.

6. The method of claim 4 , wherein the width of at least one of the histogram intervals is based on at least a maximum of a set that includes minimums of histogram intervals of the feature histograms.

7. The method of claim 1 , wherein:

the feature histogram initially has histogram intervals of equal width; and

at least one set of adjacent consistent histogram intervals is merged.

8. The method of claim 7 , wherein a set of histogram intervals is consistent if a difference between the feature histograms over the set of histogram intervals does not change sign.

9. The method of claim 7 , wherein the histogram intervals of the feature histogram are iteratively so merged until no adjacent histogram intervals are consistent.

10. The method of claim 7 , wherein at least one further set of adjacent histogram intervals is merged, the at least one further set including a histogram interval over which a minimum distance between the feature histograms occurs.

11. The method of claim 10 , wherein:

a number of histogram intervals of the feature histogram is greater than a target; and

the histogram intervals of the feature histogram are iteratively further merged until the number of histogram intervals is equal to the target.

12. The method of claim 7 , wherein a target number of histogram intervals for the feature histogram is at least an order of magnitude less than the initial number of histogram intervals.

13. The method of claim 7 , wherein a histogram interval is split so as to maximize information gain in resulting histogram intervals.

14. The method of claim 13 , wherein:

a number of histogram intervals of the feature histogram is less than a target; and

the histogram intervals of the feature histogram are iteratively so split until the number of histogram intervals is equal to the target.

15. A method for object detection, the method comprising:

a processor executing the acts of:

generating a feature space histogram for each feature extractor and object class by applying the feature extractor to training data labeled with the object class;

determining a number of the feature space intervals of the feature histogram;

instantiating a lookup table for a plurality of weak classifiers;

setting entries in the lookup table corresponding to the feature space intervals;

selecting a weak classifier from the plurality of weak classifiers based on, at least, Bayesian errors associated with the plurality of weak classifiers;

determining a strong classifier based on at least the selected weak classifiers; and

modifying a computer-readable medium based on at least the strong classifier.

16. The method of claim 15 , wherein selecting from the plurality of weak classifiers comprises selecting a weak classifier with a minimum associated Bayesian error.

17. The method of claim 15 , wherein:

the plurality of weak classifiers comprises a plurality of sets of weak classifiers; and

selecting from the plurality of weak classifiers comprises selecting a weak classifier from a set of the plurality of sets having a minimum associated Bayesian error in the set.

18. A system comprising:

a memory;

a processor coupled to the memory;

a Bayesian stump generator stored in the memory and executable on the processor, configured to instantiate at least one Bayesian stump, the at least one Bayesian stump constructed for each feature and object class;

an object detection and/or recognition (ODR) function generator stored in the memory and executable on the processor, configured to instantiate an object detection function based on at least one Bayesian stump, the object detection function constructed for a particular object class;

the ODR function generator stored in the memory and executable on the processor, configured to detect an object with the object detection function; and

a Bayesian Boosting ODR module stored in the memory and executable on the processor, configured to modify a computer-readable medium responsive to the detection.

19. The system of claim 18 , wherein the object detection function is optimal with respect to Bayesian error.

20. The system of claim 18 , wherein the instantiating the object detection function comprises determining a set of Bayesian stumps each with a minimum associated Bayesian error.

Description

Automated object detection and/or recognition (ODR) is a core facility enabling sophisticated treatment of raw data. Applications are varied, but illustrative examples include detection of physical objects (from simple geometric shapes through to geographic features, vehicles and faces) in raw static images or video, as well as detection of audio objects such as songs or voices in raw audio data. In some cases, detection (i.e., detection and/or recognition) is practically the whole application, in others, it is a small part of a much larger application.

A myriad of techniques have been developed for ODR, each with its advantages and disadvantages. However, a constant theme over time has been a demand for better efficiency as raw data sets grow ever larger. For example, it is desirable to recognize aspects of non-text media available in large public computer networks to facilitate non-text media search functionality, but it is not uncommon for corresponding raw data sets to contain items numbering in the billions. At such scales even small improvements in detection speed and accuracy can have large efficiency impacts, and it is desirable to know when a particular technique has been optimally configured in some respect.

An efficient, effective and at times superior object detection and/or recognition (ODR) function may be built from a set of Bayesian stumps. Bayesian stumps may be constructed for each feature and object class, and the ODR function may be constructed from the subset of Bayesian stumps that minimize Bayesian error for a particular object class. That is, Bayesian error may be utilized as a feature selection measure for the ODR function. Furthermore, Bayesian stumps may be efficiently implemented as lookup tables with entries corresponding to unequal intervals of feature histograms. Interval widths and entry values may be determined so as to minimize Bayesian error, yielding Bayesian stumps that are optimal in this respect.

This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

The same numbers are used throughout the disclosure and figures to reference like components and features.

In an embodiment of the invention, an efficient object detection and/or recognition (ODR) function is built from a set of Bayesian stumps. A Bayesian stump may be implemented with a lookup table (LUT) having entries corresponding to intervals of a feature histogram. Each Bayesian stump may have a specified (and possibly same) number of entries. The entries may correspond to feature intervals having unequal width. The width of the intervals may be automatically determined. The width of the intervals and the value of corresponding lookup table entries may be set so as to minimize Bayesian error. That is, the Bayesian stump may be constructed so as to be optimal with respect to Bayesian error for a given feature and lookup table.

Such Bayesian stumps may be constructed for each feature and object class. Each such Bayesian stump may be considered, by itself, a weak classifier. An adaptive boosting (“Adaboost”) technique may be utilized to construct a strong classifier, that is, an effective ODR function, based on Bayesian stumps. Such boosting techniques require a way of selecting between weak classifiers and, in an embodiment of the invention, those Bayesian stumps are selected which minimize Bayesian error. That is, Bayesian error may serve as a feature selection measure for the strong classifier, and is at times superior to conventional measures such as Bhattacharyya distance or Jensen-Shannon entropy in terms of both speed and accuracy. As used herein, the term “Bayesian boosting” includes construction of Bayesian stumps as well as construction of ODR functions based on Bayesian stumps in accordance with an embodiment of the invention.

Before describing further details of object detection in accordance with an embodiment to the invention, it will be helpful to have reference to an example computing environment suitable for incorporating such. **100**. The computing environment **100** depicts a computer **102** including a computer operating system **104**, a computing application **106**, and a graphical user interface (GUI) **108**. In this example, the application **106** includes a Bayesian boosting ODR module **110**, and the computer maintains a set of training data **112** for the Bayesian boosting ODR module **110**.

The computer **102** may be any suitable computing device. Examples of suitable computing devices include mainframes, minicomputers, desktop computers, personal computers (PCs), workstations, portable computers, laptop computers, tablet computers, personal digital assistants (PDAs), mobile telephones, programmable consumer electronics devices, routers, gateways, switches, hubs, and suitable combinations thereof.

The computer **102** may include one or more processing units capable of executing instructions to perform tasks, as well as one or more types of computer-readable media such as volatile and/or non-volatile memory capable of storing data, computer programs and/or computer program components. Such computer programs and components may include executable instructions, structured data and/or unstructured data organized into modules, routines and/or any suitable programmatic object. Such computer programs and components may be created by and/or incorporate any suitable computer programming language.

The computer **102** may include a wide variety of input/output (IPO) devices not shown in **102** include any one or more of magnetic media (such as hard disks), optical media such as compact disks (CDs) and communication media. Communication media may include any one or more of wired communication media such as copper wire, coaxial cable and optical fiber, as well as wireless communication media such as electromagnetic media including radio, microwave, infra-red and laser light. In an embodiment of the invention, each computer-readable medium is tangible.

For clarity, embodiments of the invention may be described herein with reference to symbolic operations such as those of a computer programming language. Such symbolic operations and any data that they act upon correspond to physical states of components and changes in components of computing devices such as the computer **102** in a manner well understood by one of skill in the art. In an embodiment of the invention, each such operation and its associated data may be fully implemented in hardware.

The application **106** may be any suitable application making use of ODR functionality. Although **102** including only one application **106**, each embodiment of the invention is not so limited, and the computer **102** may include any suitable number of application such as the application **106**. The application **106** may utilize the computer operating system **104** to access functionality of the computer **102** and/or access computer **102** functionality directly. A user of the computer **102** may interact with the application **106** through the graphical user interface **108** and/or any suitable user interface.

As will be apparent to one of skill in the art, the Bayesian boosting ODR module **110** need not be entirely incorporated into an application such as the application **106**. For example, Bayesian boosting ODR module **110** functionality may be made available to the application **106** utilizing any suitable distributed functionality technique. Similarly, the training data **112** need not be located entirely at the computer **102**, but may be suitably distributed, for example, in a network of computers such as the computer **102**.

The training data **112** may include a set S of classified examples {(x_{i}, y_{i})}, i=1, . . . , N, where x_{i }is a unit (e.g., a file or record) of raw data containing a positive or negative example (e.g., presence or absence) of a particular object and/or class of objects, y_{i }is an object classification label corresponding to a classification of the example x_{i}, and N is any suitable number of examples. The object classification labels y_{i }are taken from a set of object classification labels {w_{c}}, c=1, . . . , C, where again, C may be any suitable number. As an illustrative example that will be elaborated below, the raw data x_{i }may be static images, some of which include human faces, the set of object classification labels may be {w_{1}, w_{2}} where w_{1 }corresponds to a classification of “contains a face,” and w_{2 }corresponds to a classification of “does not contain a face,” and the set S is a set of appropriately labeled static images.

The training data **112** may be utilized by the Bayesian boosting ODR module **110** to build Bayesian stumps and an ODR function capable of detecting object classes {w_{c}}. **202** in accordance with an embodiment of the invention. The Bayesian boosting ODR module **202** may include a set of object classes **204** corresponding to the set of object classification labels {w_{c}}. The object classes **204** may determine the classes of object which the Bayesian boosting ODR module **202** is capable of discriminating between.

The Bayesian boosting ODR module **202** may further include a set of object feature extractors **206**. The set of object feature extractors **206** may be a set of functions {φ_{j}}, j=1, . . . , M, where M is any suitable number, capable of extracting some feature of the raw data, for example, capable of evaluating a feature score for a particular item of raw data. Such feature extractors φ_{j }may be linear (e.g., wavelet transforms) or non-linear (e.g., local binary pattern filters). By using the object feature extractors **206** on the training data **112** (**208** (i.e., Bayesian histograms) may be constructed for each of the object classes **204**. The feature space histograms **208** may be quantized according to a set of feature space intervals **210** which need not have equal width.

A Bayesian stump generator **212** may construct the feature space histograms **208** from the training data **112** (**210** widths, and generate one or more Bayesian stumps **214**. For example, the Bayesian stumps **214** may include weak classifiers implemented as lookup tables (LUTs) having entries corresponding to feature space intervals **210**. Such lookup tables may have a fixed and/or specified number of entries K, and the feature space intervals **210** to which the entries correspond need not have equal width. That is, the Bayesian stumps **214** may include K-bin LUT-type weak classifiers with unequal bin width.

From the Bayesian stumps **214** and an adaptive set of classifier weights **216**, an ODR function generator **218** may generate one or more ODR functions **220** capable of detecting objects belonging to one or more of the object classes **204** in raw data. For example, a particular one of the ODR functions **220** may be able to detect faces in static images.

Having described structural aspects of the Bayesian boosting ODR module **110** (**110** functionality.

At step **302**, one or more Bayesian stumps may be instantiated. For example, the Bayesian stump generator **212** (**214** based on the feature space histograms **208**. Example steps for performing step **302** are described below in more detail with reference to **304**, an ODR function may be instantiated based on the Bayesian stumps instantiated at step **302**. For example, the ODR function generator **218** may instantiate the ODR function **220** based on the Bayesian stumps **214**. Example steps for performing step **304** are described in more detail below with reference to **302** and **304** are grouped with dashed line **306** to indicate that these steps **302** and **304** are preparatory to object detection, and involve training or “learning” with the training data **112** (

At step **308**, a candidate data set may be acquired. For example, the computer **102** (**102** is connected. At step **310**, the ODR function instantiated at step **304** may be utilized to detect one or more objects in the candidate data set acquired at step **308**. Such detection may be considered to be dependent on the ODR function and/or one or more of the Bayesian stumps instantiated at step **302**. For example, the application **106** may apply the ODR function **220** (**110** apply the ODR function **220** to the candidate data set. At step **312**, an indication of objects detected at step **310** may be presented to a user of the computer **102**. For example, the application **106** may utilize the graphical user interface **108** to present such indication. In an embodiment of the invention. such indication includes modification of one or more computer-readable media.

The Bayesian stump generator **212** (**214** from the feature space histograms **208**. As part of Bayesian stump **214** generation, the Bayesian stump generator **212** may determine feature space intervals **210** and corresponding lookup table entry values that minimize Bayesian error with respect to feature space histograms **208** for complementary object classes **204**.

At step **402**, equal-interval feature space histograms may be instantiated. For example, the Bayesian stump generator **212** (_{j}(x), w_{c}) for each feature extractor φ_{j }and object class w_{c }by applying the feature extractor φ_{j }to those items of the training data **112** (_{c}. The feature space may be quantized into a number L of bins or intervals σ_{l }(e.g., 1000). The feature space intervals **210** may have a set of intervals σ_{l }for each histogram **208** corresponding to the quantization, and following step **402** the set of intervals a may contain intervals of equal width.

If the information density of the feature space is non-uniform (as is not infrequently the case) then quantizing the feature space with intervals of equal width is inefficient. A high number of intervals may be required to adequately capture details in regions of high information density and then the overhead to support the additional intervals in regions of low information density will be wasted. Instead intervals of unequal width may be utilized, but then a method of determining suitable interval widths is required. In the example depicted in **402**, and then the intervals are merged and split if the intervals satisfy particular criteria in order to achieve a target number K of intervals not necessarily of equal size.

Adjacent intervals may be examined for consistency. In an embodiment of the invention, adjacent intervals are considered consistent if an arithmetic difference between two feature histograms **208** (**208** may be two histograms for a same feature but complimentary object classes (such as w_{1 }and w_{2}). Symbolically, a region σ of adjacent intervals is consistent if:

(*f*(*x* _{r})−*g*(*x* _{r}))*(*f*(*x* _{s})−*g*(*x* _{s}))≧0

for any x_{r}, x_{s }in the region σ, where, for example, f(x)=p(φ_{j}(x), w_{1}) and g(x)=p(φ_{j}(x), w_{2}). At step **404**, adjacent consistent intervals in a set of feature intervals **210** (**404** it may be that further such merging is possible, or else it may be that there were no adjacent consistent intervals in the set to be merged. At step **406**, it may be determined if any merges occurred at step **404**. If some (i.e., more than zero) merges did occur, a procedure incorporating steps depicted in **404** to attempt to perform further such merges. Otherwise, no further merges are possible at step **404** and the procedure may progress to step **408**.

Following one or more iterations of step **404**, the number of intervals in the set may be decreased, for example, from the initial number of intervals L to a current number of intervals L′. At step **408**, the current number of intervals L′ may be compared to the target number of intervals K. If the current number of intervals L′ is still greater than the target number of intervals K, the procedure may progress to step **410**. If the current number of intervals L′ is now less than the target number of intervals K, the procedure may progress to step **412**. If the current number of intervals L′ is now equal to the target number of intervals K, the procedure may progress to step **502** of **414** on

At step **410**, the set of L′ feature intervals **210** (_{j}(x), w_{1}) and p(φ_{j}(x), w_{2}). may be calculated for each interval in the set of L′ feature intervals **210**, and the interval l* may be an interval for which a minimum such distance is calculated. Symbolically:

*l**=arg min_{l} *|p*(*l,w* _{1})−*p*(*l,w* _{2})|

where l is any of the set of L′ feature intervals **210** (**416**, the interval l* found in step **410** may be merged with the intervals to which it is adjacent. Typically, interval l* will be merged with adjacent intervals l*−1 and l*+1, however, interval l* may only be merged with one other interval if it is an end interval at one extreme of the feature space. The procedure may then progress again to step **408** to test if the target number of intervals K has been achieved.

In contrast, at step **412**, the set of L′ feature intervals **210** (_{j}(x), w_{1}) and p(φ_{j}(x), w_{2}). may be calculated for each interval in the set of L′ feature intervals **210** (i.e., the minimum of each interval may be calculated), and the interval l* may be an interval for which the calculated minimum reaches a maximum. Symbolically:

*l**=arg max_{l}[min(*p*(*l,w* _{1}), *p*(*l,w* _{2}))]

where, again, l is any of the set of L′ feature intervals **210** (**418**, the interval l* found in step **412** may be split to form two new intervals such that mutual information gain in the new intervals is maximized. Suppose the interval l* is the result of merging T subintervals, then the internal threshold k* at which to split the interval l* so as to maximize mutual information gain in the new intervals may be found by finding where I(k) reaches a maximum, where:

and k=1, . . . , T. Symbolically:

*k**=arg max_{k } *I*(*k*)

Once the desired number of intervals K is achieved, at step **502**, a lookup table having K entries may be instantiated for the corresponding Bayesian stump. For example, the Bayesian stump generator **212** (**214**. The lookup table entries may then be set with steps **504**, **506** and **508**.

At step **504**, a next interval k of the K intervals may be selected (or else the first of the K intervals if none has previously

been selected). At step **506**, the corresponding entry of the lookup table instantiated at step **502** may be set so as to minimize Bayesian error. For example, the entry of the lookup table may be set to w_{1 }if that choice minimizes Bayesian error (i.e., P(w_{1}|x)), or otherwise w_{2 }(if P(w_{2}|×)<P(w_{1}|x)). The lookup table entry may be set to the object class label (e.g., the value of w_{1 }or w_{2}). Alternatively, the lookup table entry may be set to the corresponding log-likelihood of the classification.

At step **508**, it may be determined if there are more intervals of the K intervals to be selected. If there are more intervals, the procedure may return to step **504** to select the next interval k. Otherwise, the procedure may progress to steps other than those depicted in _{j }having K entries corresponding to unequal intervals of the feature space.

An effective ODR function may be constructed from such Bayesian stumps. For example, the ODR function may be constructed from T such Bayesian stumps. **602**, a set of classifier weights **216** (_{i} ^{t}} may have N×T weights that influence the effect of the N examples in the training data **112** (_{i}=w_{1 }then w_{i} ^{1}=p(w_{1})/|w_{1}|, or if y_{i}=w_{2 }then w_{i} ^{1}=p(w_{2})/|w_{2}|, for each i=1, . . . , N. At step **604**, t (which will run from 1 to T) may be incremented or initialized to 1. At step **606**, the subset of classifier weights {w_{i} ^{t}} for the current value of t may be normalized. For example, the subset of classifier weights {w_{i} ^{t}} may be normalized in accordance with a probability distribution.

At step **608**, a next feature j (i.e., a next feature extractor φ_{j}) may be selected (or else a first such feature if none has been previously selected). At step **610**, a Bayesian stump w_{i} ^{t }may be trained for the feature j and the classifier weights {w_{i} ^{t}}. For example, the Bayesian stump generator **212** may instantiate a new one of the Bayesian stumps **214** for the feature j, and where the importance of each example in the feature space x′_{i}=φ_{j}(x_{i}) may be modified based on the classifier weight w_{i} ^{t }(e.g., x′_{i}=w_{i} ^{t }φ_{j}(x_{i})). At step **612**, it may be determined if there are more features to be selected. If there are more features to be selected, a procedure incorporating steps depicted in **608**. Otherwise, the procedure may progress to step **702** of **614** and **616** are diagrammatic connectors connecting

Steps **608**, **610** and **612** may result in the instantiation of M Bayesian stumps, that is, one for each object feature extractor **206** (_{t}(x) for each boosting iteration (i.e., for each t) to incorporate into the ODR function **220**. In an embodiment of the invention, the best Bayesian stump for each boosting iteration is the Bayesian stump with minimum Bayesian error, where the Bayesian error for the Bayesian stump may be calculated as a sum of the Bayesian error for each of the K lookup table entries (and thus for the corresponding feature space intervals). At step **702**, a Bayesian stump h_{t}(x) of the set of Bayesian stumps {h_{t} ^{j}(x)} instantiated at step **610** with a minimum Bayesian error may be selected for incorporation into the ODR function **220**. That is, Bayesian error may be utilized as a feature selection measure for the ODR function **220**.

At step **704**, classifier weights for the next boosting iteration (i.e., iteration t+1) may be updated based on the Bayesian stump selected at step **702**. For example, classifier weights w_{i} ^{t+1 }may be updated according to the formula: w_{i} ^{t+1}=w_{i} ^{t }exp(−h_{t}(x_{i})) for each i=1, . . . , N. At step **706**, it may be determined if there are more boosting iterations to complete (i.e., if t<T). If there are more boosting iterations to complete, the procedure may return to step **604**. Otherwise, the procedure may progress to step **708**.

At step **708**, an ODR function **220** (_{t}(x)} for t=1, . . . , T. For example, the ODR function **220** may be instantiated as a strong classifier returning the sign of a sum of the selected Bayesian stumps {h_{t}(x)}. Symbolically:

where H(x) is an ODR function **220** (

All references, including publications, patent applications, and patents, cited herein are hereby incorporated by reference to the same extent as if each reference were individually and specifically indicated to be incorporated by reference and/or were set forth in its entirety herein.

The use of the terms “a” and “an” and “the” and similar referents in the specification and in the following claims are to be construed to cover both the singular and the plural, unless otherwise indicated herein or clearly contradicted by context. The terms “having,” “including,” “containing” and similar referents in the specification and in the following claims are to be construed as open-ended terms (e.g., meaning “including, but not limited to,”) unless otherwise noted. Recitation of ranges of values herein are merely indented to serve as a shorthand method of referring individually to each separate value inclusively falling within the range, unless otherwise indicated herein, and each separate value is incorporated into the specification as if it were individually recited herein. All methods described herein can be performed in any suitable order unless otherwise indicated herein or clearly contradicted by context. The use of any and all examples, or exemplary language (e.g., “such as”) provided herein, is intended merely to better illuminate embodiments of the invention and does not pose a limitation to the scope of the invention unless otherwise claimed. No language in the specification should be construed as indicating any non-claimed element as essential to an embodiment of the invention.

Preferred embodiments of the invention are described herein, including the best mode known to the inventors for carrying out the invention. Variations of those preferred embodiments may become apparent to those of ordinary skill in the art upon reading the specification. The inventors expect skilled artisans to employ such variations as appropriate, and the inventors intend for the invention to be practiced otherwise than as explicitly described herein. Accordingly, embodiments of the invention include all modifications and equivalents of the subject matter recited in the following claims as permitted by applicable law.

Patent Citations

Cited Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|

US6954544 | May 23, 2002 | Oct 11, 2005 | Xerox Corporation | Visual motion analysis method for detecting arbitrary numbers of moving objects in image sequences |

US7020337 * | Jul 22, 2002 | Mar 28, 2006 | Mitsubishi Electric Research Laboratories, Inc. | System and method for detecting objects in images |

US7162076 | Jan 27, 2004 | Jan 9, 2007 | New Jersey Institute Of Technology | Face detection method and apparatus |

US7194114 * | Oct 7, 2002 | Mar 20, 2007 | Carnegie Mellon University | Object finder for two-dimensional images, and system for determining a set of sub-classifiers composing an object finder |

US7668346 * | Mar 21, 2006 | Feb 23, 2010 | Microsoft Corporation | Joint boosting feature selection for robust face recognition |

US7720284 * | Sep 8, 2006 | May 18, 2010 | Omron Corporation | Method for outlining and aligning a face in face processing of an image |

US20040179719 | Mar 12, 2003 | Sep 16, 2004 | Eastman Kodak Company | Method and system for face detection in digital images |

US20050036690 | Aug 15, 2003 | Feb 17, 2005 | Yi Zhou | Unified bayesian framework for shape registration |

US20050264658 | May 5, 2005 | Dec 1, 2005 | Ray Lawrence A | Face detecting camera and method |

US20060088207 | Oct 22, 2004 | Apr 27, 2006 | Henry Schneiderman | Object recognizer and detector for two-dimensional images using bayesian network based classifier |

US20060233422 | May 5, 2006 | Oct 19, 2006 | Microsoft Corporation | Machine vision system and method for estimating and tracking facial pose |

US20070036414 | Aug 10, 2006 | Feb 15, 2007 | Siemens Corporate Research Inc | Method for database guided simultaneous multi slice object detection in three dimensional volumetric data |

US20080107341 * | Nov 2, 2006 | May 8, 2008 | Juwei Lu | Method And Apparatus For Detecting Faces In Digital Images |

US20080219558 * | Mar 9, 2007 | Sep 11, 2008 | Juwei Lu | Adaptive Scanning for Performance Enhancement in Image Detection Systems |

WO2005073896A1 | Jan 31, 2005 | Aug 11, 2005 | Koninklijke Philips Electronics, N.V. | Continous face recognition with online learning |

Non-Patent Citations

Reference | ||
---|---|---|

1 | Birge et al., "How many bins should be put in a regular histogram", pre-publication copy, presented at the University of Paris on Apr. 2002, 22 pages, published in ESAIM: P&S, Feb. 2006, vol. 10, pp. 24-45. | |

2 | Bourdev et al., "Robust Object Detection Via Soft Cascade", Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05), San Diego, California, Jun. 20-26, 2005, vol. 2, pp. 236-253. | |

3 | Brubacker et al., "Towards Optimal Training of Cascade Detectors", 9th European Conference on Computer Vision (ECCV'06), Graz, Austria, May 7-13, 2006, 13 pages. | |

4 | Huang et al., "Boosting Nested Cascade Detector for Multi-View Face Detection", 17th International Conference on Pattern Recognition (ICPR'04), Cambridge, United Kingdom, Aug. 23-26, 2004, vol. 2, pp. 415-418. | |

5 | Huang et al., "Jensen-Shannon Boosting Learning for Object Recognition", IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05), San Diego, California, Jun. 20-26, 2005, vol. 2, pp. 144-149. | |

6 | Huang, et al., "Jensen-Shannon Boosting Learning for Object Recognition", IEEE, 2005, pp. 1-6. | |

7 | Javed et al., "Online Detection and Classification of Moving Objects Using Progressively Improving Detectors", IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05), San Diego, California, Jun. 20-25, 2005, vol. 1, pp. 696-701. | |

8 | Javed, et al., "Online Detection and Classification of Moving Objects Using Progressively Improving Detectors", retrieved on Apr. 17, 2007, at >, IEEE, 2006, pp. 1. | |

9 | Javed, et al., "Online Detection and Classification of Moving Objects Using Progressively Improving Detectors", retrieved on Apr. 17, 2007, at <<http://ieeexplore.ieee.org/search/wrapper.jsp?arnumber=1467336>>, IEEE, 2006, pp. 1. | |

10 | Leinhart et al., "An Extended Set of Haar-like Features for Rapid Object Detection", IEEE International Conference on Image Processing (ICIP'02), Rochester, NY, Sep. 22-25, 2002, vol. 1, pp. 900-903. | |

11 | Levi et al., "Learning Object Detection From a Small Number of Examples: The Importance of Good Features", IEEE Computer Science Conference on Computer Vision and Pattern Recognition (CVPR'04), Washington, DC, Jun. 27-Jul. 2, 2004, vol. 2, pp. 43-60. | |

12 | Levi, et al., "Learning Object Detection From a Small Number of Examples: The Importance of Good Features", IEEE, 2004, pp. 1-8. | |

13 | Li, et al., "Boosting Local Feature Based Classifiers for Face Recognition", retrieved on Apr. 17, 2007, >, IEEE, 2006, pp. 1. | |

14 | Li, et al., "Boosting Local Feature Based Classifiers for Face Recognition", retrieved on Apr. 17, 2007, <<http://ieeexplore.ieee.org/xpl/freeabs—all.jsp?arnumber=1384880>>, IEEE, 2006, pp. 1. | |

15 | Luo, "Optimization Design of Cascaded Classifiers", IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05), San Diego, California, Jun. 20-26, 2005, vol. 1, pp. 480-485. | |

16 | Luo, "Optimization Design of Cascaded Classifiers", IEEE, 2005, pp. 1-6. | |

17 | Nevatia, "Detection of Multiple, Partially Occluded Humans in a Single Image by Bayesian Combination of Edgelet Part Detectors", retrieved on Apr. 17, 2007, at >, IEEE, 2006, pp. 1. | |

18 | Nevatia, "Detection of Multiple, Partially Occluded Humans in a Single Image by Bayesian Combination of Edgelet Part Detectors", retrieved on Apr. 17, 2007, at <<http://ieeexplore.ieee.org/xpl/freeabs—all.jsp?arnumber=1541243>>, IEEE, 2006, pp. 1. | |

19 | Osuna et al., "Training Support Vector Machines: An Application to Face Detection", Conference on Computer Vision and Pattern Recognition (CVPR'97), San Juan, Puerto Rico, Jun. 17-19, 1997, 8 pages. | |

20 | Osuna, et al., "Training Support Vector Machines: An Application to Face Detection", Proceeding of CVPR, 1997, pp. 1-8. | |

21 | Rissanen et al., "Density Estimation by Stochastic Complexity", IEEE Transactions on Information Theory, vol. 38, No. 2, Mar. 1992, pp. 315-323. | |

22 | Rissanen, et al., "Density Estimation by Stochastic Complexity", IEEE, vol. 38, No. 2, 1992, pp. 315-323. | |

23 | Rowley et al., "Neural Networked-Based Face Detection", IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 20, No. 1, Jan. 1998, pp. 23-38. | |

24 | Rowley, et al., "Neural Networked-Based Face Detection", IEEE, vol. 20, No. 1, 1998, pp. 23-38. | |

25 | Schapire et al., "Improved Boosting Algorithms Using Confidence-rated Predictions", Machine Learning, vol. 37, No. 3, 1999, pp. 297-336. | |

26 | Schneiderman, "Learning a Restricted Bayesian Network for Object Detection", IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'04), vol. 2, Washington, DC, Jun. 27-Jul. 2, 2004, 8 pages. | |

27 | Schneiderman, "Learning a Restricted Bayesian Networked for Object Detection", IEEE, 2004, pp. 1-8. | |

28 | Sun et al., "Automatic Cascade Training with Perturbation Bias", IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'04), vol. 2, Washington, DC, Jun. 27-Jul. 2, 2004, 8 pages. | |

29 | Sun, et al., "Automatic Cascade Training with Perturbation Bias", IEEE, 2004, pp. 1-8. | |

30 | Viola et al., "Rapid Object Detection Using a Boosted Cascade of Simple Features", IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'01), vol. 1, Kauai, Hawaii, Dec. 8-14, 2001, 9 pages. | |

31 | Viola, et al., "Rapid Object Detection Using a Boosted Cascade of Simple Features", Computer Vision and Pattern Recognition, 2001, pp. 1-9. | |

32 | Viola, et al., "Robust Real-time Object Detection", Second International Workshop on Statistical and Computational Theories of Vision-Modeling, Learning, Computing, and Sampling, Vancouver, Canada, Jul. 13, 2001, 25 pages. | |

33 | Viola, et al., "Robust Real-time Object Detection", Second International Workshop on Statistical and Computational Theories of Vision—Modeling, Learning, Computing, and Sampling, Vancouver, Canada, Jul. 13, 2001, 25 pages. | |

34 | Wu et al., Abstract: "Detection of Multiple, Partially Occluded Humans in a Single Image by Bayesian Combination of Edgelet Part Detectors", Tenth IEEE International Conference on Computer Vision, vol. 1, Oct. 2005, 1 page, retrieved on Apr. 17, 2007, at >. | |

35 | Wu et al., Abstract: "Detection of Multiple, Partially Occluded Humans in a Single Image by Bayesian Combination of Edgelet Part Detectors", Tenth IEEE International Conference on Computer Vision, vol. 1, Oct. 2005, 1 page, retrieved on Apr. 17, 2007, at <<http://ieeexplore.ieee.org/xpl/freeabs—all.jsp?arnumber=1541243>>. | |

36 | Xiao et al., "Boosting Chain Learning for Object Detection", Ninth IEEE International Conference on Computer Vision (ICCV'03), vol. 1, Nice, France, Oct. 13-16, 2003, 7 pages. | |

37 | Xiao et al., "Robust Multipose Face Detection in Images", IEEE Transactiona on Circuits and Systems for Video Technology, vol. 14, No. 1, Jan. 2004, pp. 31-41. | |

38 | Xiao, et al., "Robust Multipose Face Detection in Images", IEEE, vol. 14, No. 1, 2004, pp. 31-41. | |

39 | Zhang et al., "Boosting Local Feature Based Classifiers for Face Recognition", Conference on Computer Vision and Pattern Recognition (CVPR'04), Washington, DC, Jun. 27-Jul. 2, 2004, vol. 5, pp. 87. |

Referenced by

Citing Patent | Filing date | Publication date | Applicant | Title |
---|---|---|---|---|

US8447712 * | May 21, 2013 | Evolved Machines, Inc. | Invariant object recognition | |

US8626686 | Jun 20, 2007 | Jan 7, 2014 | Evolved Machines, Inc. | Invariant object recognition |

Classifications

U.S. Classification | 706/20, 706/45, 382/156, 382/159, 382/157, 382/169, 382/160, 382/155, 706/62, 382/168, 706/46, 706/47, 382/158, 382/181, 706/48, 382/170 |

International Classification | G06F15/18 |

Cooperative Classification | G06N7/005, G06K9/6256 |

European Classification | G06N7/00P, G06K9/62B7 |

Legal Events

Date | Code | Event | Description |
---|---|---|---|

Nov 14, 2007 | AS | Assignment | Owner name: MICROSOFT CORPORATION, WASHINGTON Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:XIAO, RONG;TANG, XIAOOU;REEL/FRAME:020115/0907 Effective date: 20071107 |

Oct 28, 2014 | FPAY | Fee payment | Year of fee payment: 4 |

Dec 9, 2014 | AS | Assignment | Owner name: MICROSOFT TECHNOLOGY LICENSING, LLC, WASHINGTON Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:MICROSOFT CORPORATION;REEL/FRAME:034542/0001 Effective date: 20141014 |

Rotate